Methodology
How the work is done, in full.
This page exists so a technical reader can check us. If you are evaluating a brief, or you are the expert on the other side of one, everything you need to reproduce or challenge a finding is here and in section 14 of the document itself.
Looking for what you receive rather than how it is made? That is on the services page.
Finding candidates by jurisdiction.
Candidates are drawn from open business data filtered by industry classification and by address. Searches run at state level rather than city level, because city-name matching drops candidates wherever an administrative division and a locality name disagree, and a search that silently returns fewer results is indistinguishable from one that found fewer.
Jurisdiction is decided by the country and state recorded on the business address. For states on a national border this is load-bearing: a bounding box drawn around Washington contains British Columbia records, and one around Texas contains Mexican records. The country predicate excludes them.
Every site examined is recorded, including sites with no findings. A discovery run that reports only its successes cannot tell you its qualification rate.
- Selection
- Industry classification and business address from open data
- Granularity
- State level; city-level matching is not used
- Stop condition
- Candidate target met, or site ceiling reached
- Retention
- Every site examined is stored, findings or not
Real browsers, real viewports.
Pages are driven in Chromium under Playwright and evaluated with axe-core against WCAG 2.2 Level AA. Each page is examined at a desktop and a mobile viewport profile, and the profile is recorded against every finding, because a barrier that appears only below a breakpoint is a real barrier and needs to be attributable.
Findings are attributed to the page that rendered, not to the address requested. Where a request redirects, the evidence and the citation follow to the final page. Attribution to a pre-redirect URL is the kind of error an opposing expert finds first.
Engine and rule-set versions are captured per examination and printed in the brief. A finding that cannot be tied to a named tool version is not reproducible, and a result that cannot be reproduced is not evidence.
- Browser
- Chromium, driven by Playwright
- Rules
- axe-core, evaluated against WCAG 2.2 Level AA
- Viewports
- Desktop and mobile profiles, recorded per finding
- Redirects
- Attribution follows to the page that rendered
- Versions
- Engine and rule-set versions recorded and printed
- robots.txt
- Honored by default; opting out is an explicit act
The review record, printed in every brief.
Automated evaluation produces candidates, not findings. Results are normalized to remove duplicates and repeated instances of one underlying defect, then reviewed by an analyst who confirms or discards each one.
The brief prints all three counts. A reader can see how many candidates the tools produced, how many survived normalization, and how many a person confirmed. The shape of that funnel tells you as much about the rigor of the work as the findings do.
Where an examination produced no funnel, the section is omitted rather than printed as zeros. A row of zeros in a methodology section is a missing measurement wearing the costume of a measurement, and it is the kind of thing that ends a cross-examination badly.
Counts from a representative brief. Roughly one candidate in three survives review.
What this method does not establish.
Automated evaluation detects a minority of real accessibility barriers. A page with no detected failures has not been shown to be accessible; it has been shown to pass the checks that can be automated. We do not claim otherwise, and no brief says otherwise.
Conversely, a confirmed finding establishes that a barrier was present on a page at a recorded time. It does not establish intent, notice, harm, or that the barrier amounts to a violation of any statute. Those are legal questions and they belong to the attorney reading the document.
Full scope and coverage sets out the geographic and technical boundaries of what we examine.
Challenge any of this.
If something here does not stand up, we would rather hear it from you than from an expert witness.